摘要
推广性能是人工神经网络研究的重要方向。在推广性能的研究中,改进学习算法是提高前向网络推广性能的重要方法之一。本文对一种特殊的权值光滑BP算法的理论基础进行了仔细的研究,并将该算法首次应用于水声目标的分类问题。实验结果表明,具有权值光滑BP算法的前向网络不仅具有较光滑的连接权值,而且其推广性能也优于具有标准BPM算法的前向网络。
The problem of generalization ability of feedforward artificial neural networks (FFANN's) is one of the main obstacles in this field. It is of great interest for both theoretical analysis and practical application. The basic theory of a special kind of algorithms with smooth weights is discussed in this paper and for the first time, the algorithm is used for sonar target classification problems to improve the generalization ability of FFANN's .The results verified its effectiveness.
出处
《系统仿真学报》
CAS
CSCD
2001年第1期64-66,共3页
Journal of System Simulation
关键词
推广性能
光滑权值
BP算法
人工神经网络
feedforward artificial neural network
generalization ability: smooth weight
BP algorithm